My name is Brendan and I'm from Melbourne, Australia. I run Inflate AI, an AI Automation Agency assisting businesses in integrating custom AI solutions. I started Inflate soon after ChatGPT was released, identifying the significant impact that this technology would have on businesses. I create content to help business owners better understand this technology and assist others in learning how this impacts other businesses.
You should of done more than just 1 property in the function tool... Like name, email, inquiry, etc. Not a very helpful or realistic scenario for the video..
Hello, thanks for the feedback. Can you provide more details about how the quality was bad? I can forward this to my editor to improve future videos. Thanks!
Do you offer an option to charge for the complete setup process? I work in Business & Property loans. I am an Economist, not a techie. Any feedback would be greatly appreciated.
Great Video! Is there a way to enable this call button in the chat interface while using Voiceflow in the background? Because you chose the simple setup without voiceflow in this video. Thank you very much in advance!
Hi, other languages are possible although requires more work to create. You can prompt the entire agent in another language or prompt it to convert to another language. The biggest hurdle is to ensure the voice you are using works with the language accurately.
Another great video, Brendan, helping us all on our AI journeys. Two questions: (1) Have you switched to Zapier from Make permanently? (2) How are you arriving at a figure of 93% accuracy? A cosine similarity tool? ;-)
1. I use Zapier or Make, just depends what the clients prefers. 2. I run automated testing using AI to call the system, I will be releasing a video on this soon.
Hello Brendan, how are you? Thank you for sharing this AI Agent, we always need to update on new markets. Now, can you make this Zapier automation available to the community? Zapier lets you share templates. Thanks and success to you!
How did you come up with the pricing? Was it based on value to the client or mark up on dev costs? Seems this is in the US. How do you handle privacy notification for recording calls. Some states require the caller be notified at the beginning of the call. Otherwise it would illegal to get the transcripts.
Great video ! I think the most valuable part is when you say that you achieved 93% accuracy. As an AI engineer myself, it can't be stressed enough that any AI system cannot be accurate 100% all the time, and very few people talk about that. Also automated testing is a very good approach that I would encourage everybody building AI systems to do !
Thanks for watching. Typically we've found the most buggy areas are the platforms connected to Vapi. Elevenlabs, Deepgram, OpenAI, etc. Not a whole lot we can do when these things have downtime, although Vapi has been quite responsive in noting when these outages occur.
Banger Video, Brendan! I think people still underestimate how valuable those solutions are. Done right, I still believe that even $13k is a bargain for a tailored solution. Keep it up!
From your umpteenth test trials, which voice Ai/tts model is the best? Eleven labs? Deepgram? Something else? The one with the most human like voice and perhaps a decent latency.
Currently, I've been using ElevenLabs and DeepGram. Although, the new Real-Time API performs well in both quality and latency, although the cost is significant and has few controls over the voice.
The video is great! I'm curious, though why are you focusing solely on prompting ? Is there an advantage to using only prompts? Have you considered integrating it with a platform like Voiceflow for easier conversation management? Keep up the good work bro!
Hello, I have done the Voiceflow integration in a previous video, and I do believe there are some advantages to doing so. Although, the slight increase in latency wasn't ideal and ultimately the conversational aspect for this build wasn't too complex. Something like order management or something requiring nuanced details may be better with that setup.
isn't this system super inefficient and resource intensive? zapier zaps add up quickly and your prompt is *massive*. might recommend checking out openai swarm
The real-time API is very expensive and not nearly as controllable as the current tech stack. It sounds great and it's definitely the right direction. But for our clients, it not something we are currently recommending. Would love to hear your thoughts.
@@brendanautomation I think we should focus on making a good UX UI and focus more on the customer while we implement openai at cost price without making a profit, and when the price drops (which it will soon) we will have a profit and the best possible service
@@brendanautomation I think we should focus on making a good UX UI and focus more on the customer while we implement openai at cost price without making a profit, and when the price drops (which it will soon) we will have a profit and the best possible service
Wait, this is wild. Ya'll are connecting the TTS to an LLM? So how long until the client you are representing calls enraged when some kid jailbreaks it makes it say something awful? Or it just hallucinates details and causes the customer to become enraged at the inaccurate info? Like, I get that we're at the stage in business where everyone is okay with not really considering why we need to over-engineer something that templates and a speak-n-spell have been able to do reliably for decades, but like... there are consequences to moving fast and breaking things. I would find it really funny to read some of these generated transcripts, especially this month if grandma has the TV blaring and it picks up details about something spooky. "Customer in property xyz is experiencing a home invasion from zombies, when told to hang up and contact law enforcement immediately she refused and became preoccupied with a leaky faucet, despite the clear and imminent danger."
@@brendanautomation Do you mind explaining a bit more? I was under the impression that LLM's are considered "internal adversaries" by security researchers. A search on the ArXiv for "llm attack" reveals daily analysis of different attack vectors related to either single-shot or multi-shot attacks. I find it very dubious that simply prompting or providing guidelines can remove these clear structural issues with LLMs when fine-tuning, system prompts, RLHF, etc have been unsuccessful.
@@brendanautomation Look, there's a reason that LLM's are considered a form of insider adversary, and furthermore - if you think you can prompt away these kinds of problems, I'd invite you to search on the ArXiV for "llm attack". There are single-shot and multi-shot jailbreaks published almost daily. No amount of fine tuning, RLHF, or prompting can remove exploits that work on the attention mechanism itself. Now, for something like this - yeah, you're probably okay. but my original question is something you should test for. These systems can be led astray and at least for larger clients that's a huge problem. The last thing you want is to be the reason for a PR nightmare because of a prank, and it goes deeper than that - regulators are slowly catching up, so a system like yours likely would not pass muster for say, a medical clinic, at least not for long. And don't get me wrong! LLM's are amazing, but we should be playing to their strengths, and being very very mindful of their weaknesses. Templates can be remarkably versatile and exhaustive - and if you insist on paying for the overhead of an llm, they might also offer a defense if you're using something like function calling.
There is no "right" way to go. This setup worked best for our client, so we delivered this for them. We didn't need to hook into any CRM, so we didn't need GHL.
@@brendanautomation Hey Brendan, I noticed comments are off on this video in Skool so posting here. What do you normally include in the $500-$2k+/m maintenance package? Liek, once it's built and working, what is being done each month to justify the fee? I Want to get recurring also, but not sure how to justify this to the client. Would love to hear how you're thinking about this!
At the end of your video you mentioned what you provide, so that gave some more insight. It's basically providing insights and analytics and areas to improve the system to make it even better. I imagine at some point, the system is near perfect and they drop off the retainer?
Definitely looking forward to this one! Would love for you to drop it in the Skool community first, but I’ll wait and watch it with the regular folk 😂😂😂 People, join his community. Its awesome!
Hey Brendan, love the video (I actually watched it a while back and am just coming back to contact you lol). Tixae Agents (formerly VG) is launching an Influencer Partnership program (no cost, no contracts). If you're interested in hearing more what's the best way to contact you?
